def testTooSmall(self):
with self.assertRaises(ValueError):
param = tf.ones([1], dtype=np.float16)
checked_param = distribution_util.embed_check_categorical_event_shape(
param)
if tf.executing_eagerly(): return
with self.assertRaisesOpError('must have at least 2 events'):
param = tf1.placeholder_with_default(np.ones([1],
dtype=np.float16),
shape=None)
checked_param = distribution_util.embed_check_categorical_event_shape(
param)
self.evaluate(checked_param)
def testTooLarge(self):
with self.assertRaises(ValueError):
param = tf.ones([int(2**11+1)], dtype=tf.float16)
checked_param = distribution_util.embed_check_categorical_event_shape(
param)
if tf.executing_eagerly(): return
with self.assertRaisesOpError(
'Number of classes exceeds `dtype` precision'):
param = tf1.placeholder_with_default(
np.ones([int(2**11+1)], dtype=np.float16), shape=None)
checked_param = distribution_util.embed_check_categorical_event_shape(
param)
self.evaluate(checked_param)
def _maybe_assert_valid_concentration(self, concentration, validate_args):
"""Checks the validity of the concentration parameter."""
if not validate_args:
return concentration
concentration = distribution_util.embed_check_categorical_event_shape(
concentration)
return distribution_util.with_dependencies([
tf.compat.v1.assert_positive(
concentration, message="Concentration parameter must be positive."),
], concentration)
def __init__(self,
logits=None,
probs=None,
dtype=tf.int32,
validate_args=False,
allow_nan_stats=True,
name="Categorical"):
"""Initialize Categorical distributions using class log-probabilities.
Args:
logits: An N-D `Tensor`, `N >= 1`, representing the log probabilities
of a set of Categorical distributions. The first `N - 1` dimensions
index into a batch of independent distributions and the last dimension
represents a vector of logits for each class. Only one of `logits` or
`probs` should be passed in.
probs: An N-D `Tensor`, `N >= 1`, representing the probabilities
of a set of Categorical distributions. The first `N - 1` dimensions
index into a batch of independent distributions and the last dimension
represents a vector of probabilities for each class. Only one of
`logits` or `probs` should be passed in.
dtype: The type of the event samples (default: int32).
validate_args: Python `bool`, default `False`. When `True` distribution
parameters are checked for validity despite possibly degrading runtime
performance. When `False` invalid inputs may silently render incorrect
outputs.
allow_nan_stats: Python `bool`, default `True`. When `True`, statistics
(e.g., mean, mode, variance) use the value "`NaN`" to indicate the
result is undefined. When `False`, an exception is raised if one or
more of the statistic's batch members are undefined.
name: Python `str` name prefixed to Ops created by this class.
"""
parameters = dict(locals())
with tf.compat.v1.name_scope(name, values=[logits, probs]) as name:
self._logits, self._probs = util.get_logits_and_probs(
logits=logits,
probs=probs,
validate_args=validate_args,
multidimensional=True,
name=name)
if validate_args:
self._logits = util.embed_check_categorical_event_shape(
self._logits)
logits_shape_static = self._logits.shape.with_rank_at_least(1)
if logits_shape_static.ndims is not None:
self._batch_rank = tf.convert_to_tensor(
value=logits_shape_static.ndims - 1,
dtype=tf.int32,
name="batch_rank")
else:
with tf.compat.v1.name_scope(name="batch_rank"):
self._batch_rank = tf.rank(self._logits) - 1
logits_shape = tf.shape(input=self._logits, name="logits_shape")
num_categories = tf.compat.dimension_value(logits_shape_static[-1])
if num_categories is not None:
self._num_categories = tf.convert_to_tensor(
value=num_categories,
dtype=tf.int32,
name="num_categories")
else:
with tf.compat.v1.name_scope(name="num_categories"):
self._num_categories = logits_shape[self._batch_rank]
if logits_shape_static[:-1].is_fully_defined():
self._batch_shape_val = tf.constant(
logits_shape_static[:-1].as_list(),
dtype=tf.int32,
name="batch_shape")
else:
with tf.compat.v1.name_scope(name="batch_shape"):
self._batch_shape_val = logits_shape[:-1]
super(Categorical, self).__init__(
dtype=dtype,
reparameterization_type=reparameterization.NOT_REPARAMETERIZED,
validate_args=validate_args,
allow_nan_stats=allow_nan_stats,
parameters=parameters,
graph_parents=[self._logits, self._probs],
name=name)
def testUnsupportedDtype(self):
param = tf.convert_to_tensor(value=np.ones([2**11 + 1]).astype(
tf.qint16.as_numpy_dtype),
dtype=tf.qint16)
with self.assertRaises(TypeError):
distribution_util.embed_check_categorical_event_shape(param)
def __init__(
self,
logits=None,
probs=None,
dtype=tf.int32,
validate_args=False,
allow_nan_stats=True,
name="Categorical"):
"""Initialize Categorical distributions using class log-probabilities.
Args:
logits: An N-D `Tensor`, `N >= 1`, representing the log probabilities
of a set of Categorical distributions. The first `N - 1` dimensions
index into a batch of independent distributions and the last dimension
represents a vector of logits for each class. Only one of `logits` or
`probs` should be passed in.
probs: An N-D `Tensor`, `N >= 1`, representing the probabilities
of a set of Categorical distributions. The first `N - 1` dimensions
index into a batch of independent distributions and the last dimension
represents a vector of probabilities for each class. Only one of
`logits` or `probs` should be passed in.
dtype: The type of the event samples (default: int32).
validate_args: Python `bool`, default `False`. When `True` distribution
parameters are checked for validity despite possibly degrading runtime
performance. When `False` invalid inputs may silently render incorrect
outputs.
allow_nan_stats: Python `bool`, default `True`. When `True`, statistics
(e.g., mean, mode, variance) use the value "`NaN`" to indicate the
result is undefined. When `False`, an exception is raised if one or
more of the statistic's batch members are undefined.
name: Python `str` name prefixed to Ops created by this class.
"""
parameters = dict(locals())
with tf.name_scope(name, values=[logits, probs]) as name:
self._logits, self._probs = util.get_logits_and_probs(
logits=logits,
probs=probs,
validate_args=validate_args,
multidimensional=True,
name=name)
if validate_args:
self._logits = util.embed_check_categorical_event_shape(
self._logits)
logits_shape_static = self._logits.shape.with_rank_at_least(1)
if logits_shape_static.ndims is not None:
self._batch_rank = tf.convert_to_tensor(
logits_shape_static.ndims - 1,
dtype=tf.int32,
name="batch_rank")
else:
with tf.name_scope(name="batch_rank"):
self._batch_rank = tf.rank(self._logits) - 1
logits_shape = tf.shape(self._logits, name="logits_shape")
if logits_shape_static[-1].value is not None:
self._event_size = tf.convert_to_tensor(
logits_shape_static[-1].value,
dtype=tf.int32,
name="event_size")
else:
with tf.name_scope(name="event_size"):
self._event_size = logits_shape[self._batch_rank]
if logits_shape_static[:-1].is_fully_defined():
self._batch_shape_val = tf.constant(
logits_shape_static[:-1].as_list(),
dtype=tf.int32,
name="batch_shape")
else:
with tf.name_scope(name="batch_shape"):
self._batch_shape_val = logits_shape[:-1]
super(Categorical, self).__init__(
dtype=dtype,
reparameterization_type=reparameterization.NOT_REPARAMETERIZED,
validate_args=validate_args,
allow_nan_stats=allow_nan_stats,
parameters=parameters,
graph_parents=[self._logits,
self._probs],
name=name)
